The power of many: when genetics met yeasts and high-throughput.

Save

ABSTRACT

In recent years, complex technological capabilities have evolved, driven by the need to solve complex and integrative biological questions through global analyses. New equipment allows the scaling up and automation of processes which previously were carried out on a very limited scale. Concomitant with the availability of sophisticated technology developed to increase experimental processes in parallel, it is essential to have a versatile biological model that allows us to generate and handle large collections of genetic variants with the greatest possible capacity.

This is a critical aspect in order to approach saturation levels in each search, to minimise false‐positive and false‐negative rates. In this sense, yeasts represent a prototypical eukaryotic model with a remarkable genome editing potential: it is possible to generate and phenotype several thousand genetically diverse strains on a high‐throughput scale. Here we review, with a focus on the most recent years, the combination of the power of yeasts with the technical advances in multi‐level processing capability.

This has resulted in unprecedented global mappings of a multitude of biological processes with important implications that reshape our knowledge of fundamental biology, evolution and biotechnological/biomedical applications.

INTRODUCTION

The concept of high‐throughput (HT) refers to the ability to handle a large number of samples in parallel from a procedure that originally could only be carried out with one or a just a few samples. The increase in the processing capacity involves the opening of any bottleneck in an experimental process. For example, this may be achieved by increasing the number and genetic diversity of the strains analysed, multiplying the number of samples and/or conditions in the same space or platform, massively parallel synthesis of oligonucleotides, mass construction of plasmid libraries, multiplying the transformation processes, etc.

In addition, multiplying the experimental analysis power itself allows sufficient technical and biological replicates to provide more statistically robust results. This has become possible through the development of applied robotics, multiplying the amount of simultaneous pipetting, mixing and shaking in multi‐well plates and doing so iteratively with different plates. Another key automation aims to address the need to grow, replicate and mate thousands of yeast strains in a very limited space and time and enable the in vitro synthesis of large DNA fragments in a significantly cheaper, faster and systematic way.

In order to complete the technological pipeline of a global study, automated multivariate phenotype readers have also been developed, as well as specific software for mass analysis. These systems are able to compare experiments against controls and between different experimental replicates, automatically highlighting positions above or below a given threshold, as well as clustering the results according to the criteria of interest. Depending on the combination of technologies used, the scaling‐up factor can be tens, hundreds, thousands or even millions.

The most obvious HT paradigm, one that has paved the way for a large number of upstream experiments, is the increase in nucleic acid sequencing capacity. A PhD student in the late 1990s had to perform individual Sanger sequencing reactions for each DNA fragment and for each of its bases separately, then run the samples on huge acrylamide gels and read the results manually to achieve about 300 bases of confidence per reaction. Today that same person – a principal investigator in their late forties – is able to read hundreds of millions of bases using roughly the same amount of time and labour.

In fact, the technology behind this massive throughput has itself revolutionised the way of thinking on a multitude of genetic and biochemical questions that involve knowing DNA or RNA sequences.

Yeast models have been pioneers in the application of this HT development to provide general quantitative and qualitative transcriptome landscapes in eukaryotes by RNA sequencing (Wilhelm et al., 2008; Nagalakshmi et al., 2008), as well as fresh insights into genome variation and genotype–phenotype relationships across natural isolates (Liti et al., 2009; Jeffares et al., 2015; Peter et al., 2018). Cell‐based HT screens have undoubtedly been led by the budding yeast Saccharomyces cerevisiae followed by other evolutionary distant yeasts such as Schizosaccharomyces pombe or Candida albicans, along with other ‘non‐conventional’ yeasts. Altogether, these models have boosted our ability to draw global maps of genetic interactions in order to understand the relationships between genotype and phenotype, both in basic science and in biomedical and biotechnological applications.

Among other advantages, these models are inexpensive to grow and do not require special environmental conditions. It is possible to grow a large number of individual clones in a much reduced space. They have a short generation time, and their growth rates are directly correlated with their genetic fitness.

In genetic terms, it is possible to work with haploid and diploid strains, and to induce mating and meiotic segregation. In addition, their homology‐directed repair (HDR) pathways are very powerful (Baudin et al., 1993) and clustered regularly interspaced short palindromic repeats‐caspase (CRISPR‐Cas) editing systems work efficiently (Holland & Blazeck, 2022) on them. This allows for more than one simultaneous edit in the same cell and systematic targeted genomic editing (Dicarlo et al., 2013; Jacobs et al., 2014; reviewed for yeast and fungi in Liao et al., 2021).

Last but not least, yeasts are ideal models in the application of cutting‐edge ‐omics disciplines (Kumar & Snyder, 2001).

NON‐BIOLOGICAL ENGINEERING

So far, a countless number of devices to facilitate HT methods have been developed and marketed by different biotech companies. The exponential increase in the ability to handle reagents and live cells in parallel is based on the miniaturisation of processes and the use of multi‐well platforms that can accommodate hundreds or thousands of different samples. Together with programmable mechanical arms capable of physically moving the holders, processes can be iteratively repeated as many times as necessary according to the required sample density.

Despite the large number and variety of equipment available on the market, the vast majority are designed with varying degrees of specialisation for three types of generic capabilities. In some cases, more than one of these capacities can be found integrated or coupled together in the same equipment setting (Donati et al., 2023).

Liquid handling

This capability is used for handling reagents and live cells in liquid culture and/or for reading results. These machines are very versatile pipetting robots that can be complemented with different modules (such as stirrers, thermal cyclers, motorised magnets, etc.) to allow the automation of cellular and molecular processes for applications such as nucleic acid extractions, quantification, dilution, normalisation and cloning. Complex polymerase chain reaction (PCR)‐based processes such as library preparation for next generation sequencing (NGS), DNA synthesis and assembly, cell transformations, in vitro mutagenesis, genotyping or quantitative PCR (qPCR) can also be scaled up.

Other modules, coupled to magnet systems, enable the systematisation of processes associated with magnetic beads such as purifications or immunoprecipitations.

Colony handling

Other equipment has also been designed to work with solid media plates to inoculate, grow, rearrange, replicate and crossbreed large collections of strains and clones in spatially separated positions within arrays of 96, 192, 384, 768, 1,536, or even 6,144 different spots on a culture plate of just 150 cm2 (Bean et al., 2014). Considering that collections of all viable deletions of the most commonly used yeasts (S. cerevisiae and S. pombe) consist of less than 5000 clones, at the maximum working density the entire collection could be contained on a single plate. Even working at intermediate densities, the complete collection could be contained in just a dozen plates.

In this scenario, we can analyse a multitude of phenotypes in a multitude of genetically different backgrounds (Kuzmin et al., 2018).

Data acquisition and analysis

Depending on the experimental design, readout results can be one‐off or repeated at different time points for dynamic phenotypes (growth curves, enzyme activity, subcellular localisation, etc.). Consequently, the equipment required can be more or less complex. Among the simplest are high‐resolution imaging systems (to measure size, colour and shape of yeast colonies), spectrophotometers (to measure absorbance at different wavelengths, monitoring colorimetric differences related to biochemical activity or metabolite concentration, expression reporters or turbidity, which correlates the cell density with growth rate in liquid cultures).

More complex systems such as fluorimeters, plate readers, flow cytometers, mass spectrometers or automated microscopy systems can determine phenotypes such as morphology, size, number of septa or cell complexity, as well as the expression levels of fluorescent expression markers, intracellular structures and organelles, protein interactions or even metabolome content (Tarasov et al., 2014; Bean et al., 2014; Rallis & Bähler, 2016; Kamrad et al., 2020). Such studies are known as high‐content screens. As an example, by using this approach, Mattiazzi Usaj et al.

(2020) quantitatively analysed microscopy data from more than 16 million single cells of 5,627 mutant strains [knock‐outs and/or temperature‐sensitive (ts)] to look for alterations and factors responsible for intercellular heterogeneity in the endocytosis pathway. This work also represents an example of a neural network‐based image analysis (Mattiazzi Usaj et al., 2020).

The last step in the entire pipeline is the analysis of the results. This is conducted using a variety of specialised analysis software. In many cases, these are developed specifically for each experiment.

They consist of algorithms that are able to plot the sought phenotype at each position and time point so that they automatically pinpoint those positions that become of significant interest based on the criteria imposed by the researcher (Bean et al., 2014). In fact, the development of sufficiently powerful computer algorithms and artificial intelligence is critical in some types of experimental design, not only in the analysis but also in the early stages of the process. For example, it can be used for the systematic design of synthetic fragments for global genome editing.

Data repositories

Once the above pipeline is completed, all analysed and raw data are usually made available to the community. These repositories are publicly accessible interactive databases that have grown in parallel with computing capacity and the rise of ‘‐omics’ experimentation. The most prominent examples of yeast repositories are both the budding and fission yeast genome databases (SGD: https://yeastgenome.org/ and Pombase: https://www.pombase.org/ respectively) (Rutherford, Lera‐Ramírez & Wood, 2024; Engel et al., 2025) but others have also been developed in recent years, such as JaponicusDB (https://www.japonicusdb.org/) (Rutherford et al., 2022), Candida Genome Database (CGD: http://www.candidagenome.org/) (Lew‐Smith, Binkley & Sherlock, 2025), fungiDB (https://fungidb.org/fungidb/app) (Basenko et al., 2024) and other genomic repositories for systems biology (reviewed in Sharma, 2016).

These web resources provide invaluable biological information facilitated by user‐friendly interfaces for advanced search. They also provide powerful integrative analysis tools across all eukaryotes by linking to broader databases, empowering users to uncover functional and structural relationships including sequence comparison (BLAST: https://blast.ncbi.nlm.nih.gov/Blast.cgi), gene ontology (https://geneontology.org/), protein functional analysis (InterPro: https://www.ebi.ac.uk/interpro/; UniProt: https://www.uniprot.org/), 3D structure (PDB: https://www.wwpdb.org/; AlphaFold: https://alphafold.ebi.ac.uk/), genetic and physical interactions (STRING: https://string-db.org/; Biogrid: https://thebiogrid.org/), and chemical entities of biological interest (ChEBI: https://www.ebi.ac.uk/chebi/). The integrated use of these databases may be of massive assistance for the global research network – including non‐yeast researchers – to make sense of their own data or to generate new hypotheses.

BIOLOGICAL ENGINEERING

Genome editing

In HT experiments, the number of genetic variants to work with determines the scale and density of the experimental process, and in some cases is the limiting variable. The production of genetic variants among which the search of interest is performed may be driven in a random or targeted manner (Vidan & Snyder, 2001).

Classical random mutagenesis can be achieved with chemical alkylating agents, such as methyl methanesulfonate (MMS), nitrosourea (NU), ethyl methanesulfonate (EMS), N‐ethyl‐N‐nitrosourea (ENU) etc., or physical agents, such as radiation, that produce single nucleotide changes or small insertions and deletions (indels) (Lee et al., 1992; Winston, 2008; Hoyos‐Manchado et al., 2020). These classical methods generate mutations at multiple genomic sites simultaneously and the loci affected are different from cell to cell (Fig. 1A).

Another traditional method for producing genomic variation involves biological agents such as insertion‐producing transposons. These are DNA fragments containing a selection marker flanked by particular sequences, which are recognised by an ectopically expressed transposase. This enzyme produces quasi‐random insertions into one or more sites of a cell's genome; although the landing sites can be identified with relative ease (reviewed in van Opijnen & Levin, 2020) (Fig.

1B). Recently, an ingenious method (named OrthoRep) has been developed in budding yeast to generate random mutant variants in vivo, but targeted exclusively to a chosen gene without affecting the rest of the genome. These cycles of cumulative mutagenesis can be propagated in a multitude of clonal lines in parallel so that they can be easily integrated into HT screening systems (Ravikumar et al., 2018) (for relevance to evolutionary research, see Section IV.2).

As opposed to random strategies, targeted genome editing is based on HDR systems. In several yeast models, this process occurs spontaneously at a selectable frequency. This has put yeast at the forefront of targeted genomic manipulation because, unlike in most other eukaryotic models, it is sufficient to introduce a DNA fragment into the cell with the desired modification, a selection marker and flanking sequences from the locus of interest (Fig.

1C). Nevertheless, the discovery and implementation of CRISPR‐Cas systems in all biological models has revolutionised the ability to edit metazoan genomes with unparalleled efficiency (Doudna & Charpentier, 2014; Dias et al., 2021) and it has further enhanced and diversified the potential for systematic editing in yeast (Dicarlo et al., 2013; Jacobs et al., 2014; reviewed for yeast and fungi in Liao et al., 2021). The use of Cas‐nucleases, which generate site‐specific double‐strand breaks in the genome, allows for remarkable potentialities.

On the one hand, if no template for break repair is provided, the cell utilises the non‐homologous end joining repair system, yielding indels. On the other hand, providing an edited template for the desired modification, flanked by homologous upstream and downstream ends of the cutting site, introduces scarless modifications at efficiencies over 90% in some cases (Fig. 1D).

This high efficiency circumvents the use of selection markers. This in turn avoids possible interference of marker expression with the regulation of adjacent genomic regions (Egorov et al., 2021). A further improvement, using a mutant nuclease that cuts only one of the strands (nickase), drastically decreases off‐target effects and the generation of indels when they are not desired (Satomura et al., 2017).

Another important contribution of this technology for yeast genome editing is that we can generate two or more independent genomic modifications simultaneously within the same cell with high efficiency (Bao et al., 2015; Zhao & Boeke, 2020). Therefore, multiple mutants can be obtained without having to make genetic crosses and select the meiotic product of interest. In terms of genome‐wide scale‐up of CRISPR‐Cas‐based editing, the fundamental limitation lies in the mass production of guide RNAs (gRNAs) and template fragments for homologous recombination.

However, by massively fabricating on‐chip oligonucleotides containing the gRNA and the homologous recombination template, libraries of more than 20,000 plasmids capable of editing 98% of the open reading frames (ORFs) in a single experimental process have been achieved (Bao et al., 2018). Finally, the CRISPR explosion has not only benefited traditional yeast models, but has also opened the door to systematic genome editing in other non‐conventional yeasts of biotechnological and biomedical interest, where classical editing systems do not work for various reasons. Such is the case for Scheffersomyces stipitis, Ogataea polymorpha, Ogataea thermomethanolic, Pichia pastoris, Rhodotorula toruloides, Yarrowia lipolytica, Kluyveromyces marxianusen, Kluyveromyces lactis, etc.

(Shan, Dai & Wang, 2021; Strucko et al., 2021).

For both transposon library synthesis and targeted genome editing cassettes, small non‐redundant artificial DNA sequences between 6 and 25 nucleotides, known as barcodes, can be added. After reading them by microarrays or HT sequencing, these barcodes identify unambiguously each template fragment/transposon (in a similar way to the barcodes that identify each product in a supermarket) and therefore each strain obtained from them. Another advantage of using these identity cards is that they allow working in the same flask/well/tube by mixing the members of a collection in a pool (see Section III.2.b); thus reducing the time, space and reagents of downstream experiments.

The co‐evolution of technologies for the massive systematic synthesis of oligonucleotides and the assembly of DNA fragments in vitro enables the design and construction of libraries of thousands of plasmids and/or linear fragments on demand in a reasonable time and at a reasonable cost. This allows customised expression libraries that can be arrayed for global studies of episomal gene complementation and suppression, deep mutational analysis, protein tagging, over‐expression or protein–protein interactions (Storey et al., 2019) (Fig. 1E).

Systematic mutant collections

With the tools described above, a large number of reference genetic variant collections have been generated for global analysis. Depending on the method used to identify each of their members, these collections can be kept in arrays or in strain pools.

Arrayed

In these collections, each variant is generated in a targeted manner at each locus of interest and is arranged in a matrix in which we know the genetic modification at each position. The classic example of these collections are systematic deletions of all ORFs, with these being replaced by dominant markers. Despite the automation of some steps, the first genome‐wide systematic collections in yeast required a large workload and joint efforts of laboratory consortia.

A pioneering collection of budding yeast was created to study systematically the lack of function of every gene in a eukaryotic genome (Winzeler et al., 1999; Giaever et al., 2002). It was later extended to other model yeasts such as S. pombe (Kim et al., 2010) and C. albicans (Noble et al., 2010) (Table 1). Using diploid backgrounds, it is possible to obtain heterozygous ORF deletions.

Phenotypic analysis of these diploids makes it possible to relate gene dosage to fitness and identify those loci that generate haploinsufficiency (a deleterious phenotype when one copy of the gene is functional and the other is not) (Uhl et al., 2003; Moris et al., 2016). These heterozygous genetic backgrounds also allow for transformation or mutagenesis and subsequent segregation to identify extragenic suppressors by mutation or overexpression (Szamecz et al., 2014). By inducing meiotic segregation to study each deletion in haploidy, the first large‐scale ‘essentiality’ maps could be drawn.

The labels ‘essential’, ‘slow growers’ and ‘non‐essential or dispensable’ were assigned when the lack of a gene abolishes, decreases or has no effect on proliferation, respectively. A consideration in the use of these collections is the following observation from ribosome profiling, reverse transcription (RT)‐qPCR and reporter expression: approximately 20% of deletions cause greater than twofold alterations in transcription and translation efficiency, affecting the protein abundance of their neighbouring genes (Egorov et al., 2021). Likewise, DNA microarray expression profiling assays reveal that approximately 8% of gene deletions in S. cerevisiae are associated with aneuploidy, even though the reported function of the deleted gene is not directly related to genomic stability (Hughes et al., 2000).

Therefore, it is important to take into account a possible ‘near‐neighbour’ and/or aneuploidy effects in the interpretation of the phenotypes.

In addition to the deletion of coding regions, systematic collections of deletions of non‐coding elements, such as long intergenenic non‐coding RNAs (lincRNAs), are already available (Table 1). Using CRISPR‐Cas9 editing in S. pombe, Rodriguez‐Lopez et al. (2022) recently deleted 141 and overexpressed 113 lincRNAs and they assessed viability, fitness, microscopy and fluorescence‐activated cell sorting (FACS) phenotypes for 135 different conditions.

Remarkably, about half of these lincRNAs may have a biological function, as their deletion gives rise to one or more phenotypes (Rodriguez‐Lopez et al., 2022). There are also other collections that do not involve deletions but modify the gene product in some way, such as fluorescent fusions or barcoded ts alleles linked to a dominant selection marker (Huh et al., 2003; Ben‐Aroya et al., 2008; Li et al., 2011; Kofoed et al., 2015) (see Table 1).

Pooled

If the collection is barcoded, it is possible to generate and maintain thousands of variants with different modifications in a single pool within the same tube or well. This provides an operational advantage for storage and experimental handling. Transformation with episomal vectors and linear fragments, drug exposure and competition experiments can be done with thousands of strains in the same tube or well, thus multiplying the parallel processing capacity (Kofoed et al., 2015).

One particular type of these collections that is of great interest are expression collections, which can either be episomal (from plasmids) or created by integrating regulatable and powerful exogenous promoters upstream of each genomic element of interest. These gene libraries/collections have allowed, for example, the identification of extragenic complementing functions, negative dominance or the systematic subcellular visualisation of fluorescent fusions in order to define the compartmentalisation and dynamics of each protein (Table 1; Huh et al., 2003; Mnaimneh et al., 2004; Matsuyama et al., 2006; Sopko et al., 2006). This range of possibilities opens the door to the generation of global on‐demand collections, many of which are already commercially available (Table 1).

Random collections

Random collections consist of strains with genetic modifications generated by chance along the genome or at a particular locus. As mentioned above, these may be produced by physical [ultraviolet (UV), X‐rays, etc.], chemical (mainly alkylating agents) or genetic agents (transposons, CRISPR‐Cas, mutagenic PCR, orthorep) (Fig. 1).

These approaches are proving particularly fruitful in the study of polymorphisms of many human genes of interest using the yeast cell as a bioreactor (see Section V). These collections share the characteristic that the variant responsible for the phenotype of interest is identified a posteriori, but they offer several advantages. First, in contrast to deletions, which remove the entire element, these agents typically cause single base changes or small indels.

This usually affects surgically specific motifs of a protein or may inactivate a particular domain when the protein possesses more than one. All of this reduces the likelihood of uncoupling the possible complexes where the protein might work. Furthermore, when achieving high mutation rates in a gene of interest fused to a fluorescent protein in an automated microscopy setting, it is possible to distinguish missense (fluorescent signal) from nonsense (no fluorescence) mutations within hundreds or thousands of random variants associated to a phenotype.

Second, random mutations can modify any genome's functional element that was not considered a priori. Third, single‐base changes are likely to obtain a gain of function, as well as conditional versions to work with switchable alleles of essential genes.

With the development of NGS techniques, classical limitations to identifying genetic variants responsible for a phenotype of interest have been overcome. These classical limitations occurred when: (i) the phenotype of interest is generated by two or more modifications; (ii) the mutation resides in a non‐coding site; or (iii) the phenotype of interest involves a gain of function. Whole genome sequencing identifies where the mutation/s of interest map in comparison to reference genomes (Fig.

1A,B). This makes it possible to identify tens or even hundreds of polymorphisms in parallel.

HT SCREENS APPLICATIONS

Addressing fundamental questions

To be or not to be (essential)

What are the essential and dispensable genomic elements of a eukaryotic genome? This is possibly one of the fundamental questions in biology. It was posed even before the beginning of the genomics era, and its answer is certainly not simple or absolute.

In addition to the implications in basic biology of defining a gene function as essential, in the case of yeast and fungi it is also important in applied research, such as in the search for broad‐spectrum antifungals. The possibility of analysing genomes globally has radically changed our perception – and even the definition – of the essentiality of biological functions in terms of genetic, environmental and evolutionary context (Rancati et al., 2018). Yeast models, led by the budding yeast S. cerevisiae, have been instrumental in identifying these functions.

As explained above, several methods have been developed to inactivate genome units. The first strategy to obtain a catalogue of essential genes for the survival of a cell was to delete them one by one and then to analyse whether this deletion is viable. These collections initially yielded 1105 (18.7%) and 1260 (26.1%) essential genes in budding and fission yeast respectively (Giaever et al., 2002; Kim et al., 2010).

The development of these strategies in bacteria and yeast has been progressively applied to other models including human cells in culture (Blomen et al., 2015) but was restricted to coding elements. Today, high processing capacities at all levels have led to a significant number of ingenious alternative search methods in various yeast models that unveil features and connotations of gene essentiality for any genomic elements that are otherwise inaccessible. A parallel strategy to deletions is the use of transposons as an inactivating agent for biological functions.

In principle, all elements of the genome, whether coding or non‐coding, are susceptible to inactivation. In order to assess the essentiality of an element that has been disrupted by a specific insertion, the logic behind systematic collections is reversed. In this case, transposition is induced indiscriminately in haploid cultures over many generations and the insertion sites are traced in cells that have been able to proliferate, and therefore have not been counter‐selected.

This is achieved by identifying regions adjacent to the transposon using microarrays (Kim et al., 2019) or bulk sequencing, known as transposon insertion sequencing (TIS) (van Opijnen & Levin, 2020). In high‐density insertion maps (up to one insertion every 13 base pairs), genomic regions where no disruptions are found are considered to be counter‐selected and therefore essential for life in the experimental conditions used. These experimental designs in the two most widely used model yeasts (S. cerevisiae and S. pombe) complement very well the essentiality maps previously generated by systematic deletions, identifying between 8% (S. cerevisiae) and 10% (S. pombe) of essential ORFs that had escaped systematic deletion and retained the functional gene by chromosomal rearrangements (Guo et al., 2013).

Using this strategy, new essential genomic elements have been identified beyond coding regions [including non‐coding RNAs (ncRNAs), 3′ and 5′ untranslated regions (UTRs)] and, in some cases, gene boundaries have been redefined (transcription and translation start sites) (Michel et al., 2017; Grech et al., 2019). This method has also proved very useful in Pichia pastoris (Zhu et al., 2018) and pathogenic fungi with absent or very inefficient HDR systems or with incomplete sexual cycles, such as Aspergillus fumigatus or Candida albicans, where genome editing is not straightforward. In vivo transposition in these systems is achieving qualitative steps in the identification of essential coding and non‐coding elements (Carr et al., 2010).

Segal et al. (2018) increased the catalogue of essential genes from 66 to 1610 in Candida albicans. This type of study allows for comparisons with other models.

In fact, 602 of these genes are also essential in budding and fission yeast and 130 are also described as essential in Aspergillus fumigatus, Cryptococcus neoformans, and Histoplasma capsulatum. Therefore, these elements without orthologues in the host organism may be good targets for new broad‐spectrum antifungals (Segal et al., 2018).

Non‐essential domains within essential genes

Are there non‐essential regions within an essential gene? Regardless of whether to consider a gene solely as the ORF that is translated into protein or to consider a gene as a complete transcriptional unit (including the promoter, ORF, and the 3′/5′ UTRs), extraordinarily dense transposon insertional maps show that, even within an essential gene, there may be dispensable regions. Using this approach, Michel et al.

(2017) found a repertoire of 400 essential genes that show clear boundaries of insertions that surgically delimit dispensable domains. This occurs in genes whose complete deletion is lethal, but whose protein product has more than one functional or structural domain, some of which are dispensable and therefore accept disruption events. Such is the case with Gal10, a bifunctional metabolic enzyme (epimerase and mutarotase) whose C‐terminal mutarotase activity is dispensable when both α‐ and β‐D‐galactose are available to the cell.

There arise other examples, such as Taf3, a protein of 353 amino acids, where only the first 47 are essential. Interestingly, they also find examples where insertions within essential genes are tolerated at the 5′ end and even in intermediate domains of the ORFs. These observations have obvious relevance in structure–function studies and may reveal interesting aspects of gene evolution (Michel et al., 2017).

Complementary observations in terms of subdomain essentiality have been made in an orthogonal approach by Sadhu et al. (2018). Using CRISPR‐Cas HT editing, these researchers introduced 10 premature termination codons (PTCs) into each of the 1,034 genes annotated as essential in budding yeast.

They find that PTCs are generally deleterious when introduced more than 27 codons before a gene's end, although better conserved genes and proteins with a C‐terminal catalytic domain are less tolerant. In fact, half of the genes (517) do not support any PTC even very close to the end. By contrast, 10% of genes tolerated five or more PTCs and six of these tolerated large deletions of the 3′ end (up to two thirds in some cases).

In addition to this application, this strategy can be extended to any type of allelic variants in a highly parallel manner (Sadhu et al., 2018). Extrapolating to human biology, some protein truncated variants (PTVs) have been shown to alleviate diseases such as hypercholesterolaemia, and thus represent excellent targets in the search for inhibitors to serve as treatments (Cohen et al., 2005; Dadu & Ballantyne, 2014). Yeast may therefore offer good models by introducing PTCs systematically into humanised strains (see Section V).

Conditional essentiality

The concept of essentiality is relative for many biological functions. There are many examples where an essential gene becomes dispensable and vice versa depending on the conditions under which a cell proliferates. These conditions can be environmental, genetic or a combination of both.

One of the most illustrative examples of environmental conditionality are the mutations in genes involved in the synthesis of many metabolites (auxotrophies). If the final metabolite is available in the medium and does not need to be manufactured, the genes involved in its synthesis may be dispensable (Moat, Peters & Srb, 1959; Megnet, 1964; Hoyos‐Manchado et al., 2020). Another paradigm of conditional mutants in yeast are the temperature‐sensitive mutants: certain allelic variants of many genes (essential or not) are viable at one temperature but lethal at a higher or lower one.

The conditional environmental factor may also be the presence of certain bioactive molecules in the media. Mutations that are viable under standard conditions may become lethal in the presence of certain drugs, nutrients or stress conditions (Hoyos‐Manchado et al., 2017; Rodríguez‐López et al., 2023). Conversely, the mechanisms of genetic suppression of essentiality may in some cases be mimicked by drugs that block or enhance the function of the extragenic suppressor or complementing genes, respectively.

These effects are known as genetic–chemical interactions and represent the conceptual basis underlying massive searches for bioactive compounds for biotechnological and biomedical applications (see Section VI). Another conditional setting of gene essentiality is that of genes whose functions are important only in specific biological processes, such as differentiation. There are, for example, radical differences in the gene functions required to perform an asexual cycle compared to those required for the sexual cycle.

Using again the collection of viable deletions in S. pombe, Blyth et al. (2018) mapped 354 non‐essential functions in the mitotic cycle that are critical for successful meiosis, of which only 12% have previously been implicated in conjugation, meiosis or sporulation. Such searches have important implications for the study of human gametogenesis.

Synthetic essentiality: classic genetic baits, brand‐new fishing rods

From a genetic point of view, the lethality caused by the lack of function of one gene can be rescued by either loss‐ or gain‐of‐function of another gene (extragenic suppression). Thus, a derivative of essentiality analysis is the identification of in vivo functional interactions within and between genetic pathways since these connections reveal an impressive amount of information about the functional relationships of genes, the structure of genetic networks and the biological links between them (reviewed in Costanzo et al., 2019). An important limitation for these global studies of genetic interactions in non‐yeast model organisms is achieving a high number of combinatorial double (or multiple) mutants systematically.

However, haplotypes in yeasts can be easily combined by mating haploid strains, inducing meiosis and selecting double‐mutant offspring by dominant markers. This allows systematic synthetic genetic arrays (SGAs) in which one or more query mutant/s is crossed in an automated way to a whole collection of mutants arranged in a matrix of known positions. The fitness of the resulting double mutants is estimated by the size or growth rate of the colony they form at each position in the array (Fig.

2) (Tong et al., 2001). By repeating this procedure with an increasing number of baits, global digenic (Tong et al., 2004; Ryan et al., 2012) and even trigenic interactomes have been obtained (Rallis, Townsend & Bähler, 2017; Kuzmin et al., 2018, 2021). The genetic interaction (GI) between two gene products can be positive: a mutation at one locus improves fitness or even suppresses the lethality caused by a mutation at another locus (synthetic rescue); or negative: the combination of two mutations worsens the fitness of both mutations separately, or even becomes lethal (synthetic lethality) (Fig.

2). The synthetic lethal interactions among essential genes often occur between genes within the same protein complex, or between genes in different protein complexes but within the same biological process or subcellular compartment. Positive GIs reflect more general regulatory connections related to protein homeostasis and cell‐cycle kinetics.

However, the most extreme case of positive genetic interactions – where a mutation in one gene suppresses lethality by the lack of another (extragenic suppression) – occurs mostly between genes that have a very close functional relationship. Consequently, genes whose products function within the same pathway or complex often show a similar pattern of genetic interactions; therefore, clustering can be used to identify genes encoding a pathway or complex's components (Tong et al., 2004; Vizeacoumar et al., 2010; Elbaz‐Alon et al., 2014; Costanzo et al., 2016, 2019; Kuzmin et al., 2018, 2021).

An interesting comparison of the interactomes of S. cerevisiae and S. pombe, which diverged evolutionarily between 330 and 420 million years ago (Sipiczki, 2000), reveals that almost one‐third of the interactions detected by SGA are conserved in both yeasts (Dixon et al., 2008; Ryan et al., 2012). Costanzo et al. (2016) generated a global network of interactions by analysing more than 23 million double mutants (combination of pairwise viable deletions or combination of temperature‐sensitive alleles in essential genes).

In this combinatorial they identified 550,000 negative and 350,000 positive gene interactions (Costanzo et al., 2016). By integrating the information provided by these interactomes, general principles of genetic interactions emerge. It is estimated that the number of genes comprising a ‘minimal genome’ contains 3.5 times the number of genes considered essential on their own.

Essential genes are major hubs and form the basic scaffold of the global genetic interaction network.

Although the idea of genetic interaction is a classical concept, HT techniques broaden the spectrum of detection. In addition to obtaining double deletions, completely novel searches have been performed that include any genomic element, coding or non‐coding, that genetically interacts with any other element. In this case, the baits are known mutants (deletions, conditional alleles, etc.) with a particular phenotype.

Over this background, second mutations randomly generated are sought in other genomic elements that improve the deleterious phenotype (positive interactions), worsen it (negative interactions) or even generate an alternative phenotype. A recent study by Li et al. (2019) illustrates a superb example of scaling up and integrating several of these strategies to answer fundamental questions about the essentiality and architecture of numerous genetic pathways.

The authors selected all genes annotated as essential on the left arm of chromosome II of the fission yeast S. pombe. For each locus deletion, they kept the strain alive by expressing the corresponding gene from a counter‐selectable plasmid. For the 142 deletions analysed, they searched for genetic contexts either by mutation (chemical mutagenesis and transposition) or overexpression, which would make the plasmid – and therefore the function of the bait gene – dispensable.

This approach reveals some very interesting findings: they find that 27% of those genes annotated as essential are actually dispensable (bypassable) by extragenic suppression. Subunits of protein complexes tend to share the ability, or not, to be rescued. Bypassable genes reflect a higher evolutionary rate and lower conservation than non‐bypassable genes.

It is also revealed that extragenic suppressors of the same deletion are highly likely to interact physically with each other and are therefore better predictors of functional relationships than those mapped using non‐lethal mutations. This study also illustrates how the simultaneous deletion of two genes annotated as essential can suppress each other, and that the deletion of ncRNAs can override the essentiality of a coding element (Li et al., 2019). In the same conceptual line, other studies also exploring natural diversity of the same species demonstrate that this is sufficient to respond, in terms of essentiality/fitness, differently to the same gene inactivation in different isolates (Caudal et al., 2022).

It is obvious that all this knowledge will be invaluable in the understanding of some complex multifactorial human diseases where the same mutation may display higher or lower expressivity depending on the other loci's polymorphism (Deltas, 2018), or even fully reverse severe Mendelian syndromes (Chen et al., 2016). Thus, these loci may be interesting targets for treatments as personalised medicine evolves.

Adaptation and evolution

Studies of evolutionary adaptation in yeast are closely linked to the concept of genetic interaction discussed above since they generally involve a strain with a deleterious genomic alteration recovering biological fitness to a greater or lesser extent through compensatory changes elsewhere in the genome. These studies have benefited greatly from the combined development at various levels of HT technologies that make it possible to propagate large numbers of strains (and their technical and biological replicates) over many generations, in multiple environments, automating functional assays and finally identifying, by massive sequencing, the genomic modifications that compensate for fitness in each case. Consequently, over the last few years, specific strategies have been developed to address intuitive principles experimentally in adaptive and evolutionary biology that had only been theoretically modelled or tested only in prokaryotic systems.

In one such study, Szamec et al. (2014) cultured in parallel four independent replicates of 187 null mutants for various cellular and molecular functions which caused a significant decrease in fitness, relative to the wild‐type strain, under standard laboratory conditions. They followed, in total, 748 mutant populations together with 22 independent control lines for 400 generations, monitoring growth rates in liquid medium individually.

The authors observed an average increase of 5% in the fitness of the control strains, compared to an average of 23% for strains with deleterious deletions. In many cases, fitness approached that of the wild type. In addition, more deleterious mutations are more likely to be compensated for.

To identify the underlying genetic changes, Szamec et al. (2014) sequenced the complete genomes of 41 evolved lines with high fitness improvement and compared them with their 14 corresponding ancestors. The authors found a variety of genomic modifications attributable to compensatory evolution.

These included large‐scale duplications, up to entire chromosomes, and an average of six point‐mutations and 0.5 indels per evolved line. Similarly to other studies mentioned in this review, 21 of these mutations generated premature stop codons. This suggests that fitness recovery under given conditions may involve the modulation of a second gene function.

As general principles, they concluded that evolution in these lines concentrates mainly on genes related to the missing function in the ancestor. For example, when the missing gene is part of a complex, there were 10 times more mutations in the other elements of that complex than would be expected by chance. They also found that compensatory mutations beneficial to fitness are highly dependent on growth conditions, as there is substantial variation in the fitness of the same line when subjected to environments other than the environment in which it evolved.

Based on these results, the authors proposed that fitness‐decreasing but non‐lethal mutations promote the fixation of adaptive mutations elsewhere in the genome that may in fact be beneficial in other environments (Szamecz et al., 2014).

Bao et al. (2018) followed a different strategy (named CHAnGE), in which they searched for lack of functions in S. cereviasie capable of adapting growth to unfavourable conditions for the wild‐type. To do this, they designed a CRISPR‐Cas9 library of almost 25,000 plasmids, each expressing a gRNA for a specific gene and its corresponding HDR template that included an eight‐base‐pair deletion.

In this way they managed systematically to abolish the function of all ORFs. When applied iteratively, this approach has interesting implications for the study of adaptation mechanisms for quantitative and/or epistatic traits. In a first round, they found a mutant (Siz1) that is moderately resistant to furfural and used this mutant background to repeat the process to look for double mutants that are even more resistant.

In the second round they found a strain that carries a second mutation in the LCB3 gene. Interestingly, the second mutation did not appear individually in the first round because it is not resistant to furfural on its own. However, it is only able significantly to improve fitness to furfural in combination with the lack of Siz1 function.

Thus, such approaches reveal epistasis relationships with relevance to evolutionary adaptability processes not otherwise appreciated (Bao et al., 2018) (Fig. 2).

In addition to the compensatory evolution studied by improving the overall fitness of a cell, the explosion of HT systems has allowed the design of experiments aimed at asking evolutionary questions about much more defined biological processes, such as splicing. In a recent work, Schirman et al. (2021) generated a library of approximately 19,000 barcoded large oligonucleotides that they inserted as introns into the MUD1 gene of S. cerevisiae.

Therefore, this library contains a massive synthetic combinatorial pool of intron structural variants, including sequence variations in the 5′ and 3′ splicing sites (SS), variations within the branch sites, as well as variations in the distance between them. Besides these, a large number of non‐processable control intronic sequences, two‐intron combinations and native intronic structures from 10 other different yeast species are included. The splicing efficiency of each intronic variant, associated with an individual barcode, was measured by targeted RNA sequencing.

In evolutionary terms, the authors found that the loss of the U2 small nuclear RNA auxiliary factor 1 (U2AF1), which occurs in some species, is adaptively compensated by modifying the distance between the branch site and the 3′ SS. This is achieved by the depletion of alternative 3′ SS motifs near the introns' 3′ ends at the risk of generating unwanted spliced isoforms (Schirman et al., 2021).

The recent development of ingenious genetic systems also enables the study of continuous and directed evolution in a much faster way with unprecedented processing capacity. This aims to create selection for certain proteins to modulate, improve or even change their function. Ravikumar, Arrieta & Liu (2014) developed a method (named OrthoRep) to generate recurrent and cumulative mutations in vivo in any gene/s of interest without affecting the rest of the yeast genome.

The method is based on the implementation of linear K. lactis plasmids in S. cerevisiae, whose replication depends on proteins covalently linked to the 5′ end that confer specificity for a specifically engineered highly error‐prone DNA polymerase. Thus, this approach allows powerful automated HT studies of directed evolution to search for adaptation as a function of the selective pressure imposed. To illustrate the method, the authors ‘evolve’ two proteins.

In the first case, the Plasmodium falciparum dihydrofolate reductase enzyme is made resistant to the antimalarial drug pyrethamine by selective pressure. In the second, the thermophilic Thermotoga maritima HisA enzyme (phosphoribosylformimino‐5‐aminoimidazole carboxamide ribotide isomerase, HIS6 orthologue) that normally works at 80 °C evolved to function in Saccharomyces cerevisiae at mesophilic temperatures after a long period of neutral drift. In both cases, the authors show that phenotypes are dependent on the accumulation of multiple mutations (Ravikumar et al., 2018; Zhong et al., 2020).

Ageing

Cellular ageing represents the main risk factor associated with high‐social‐impact health conditions such as neurodegenerative and cardiovascular diseases and cancer. It is well known that, in addition to environmental factors, the dynamics of natural ageing processes can be accelerated or delayed depending on certain genetic polymorphisms. A cell's lifespan can refer to the number of times a given cell can divide (replicative lifespan, RLS) or how long a cell can persist in a quiescent state before dying (chronological lifespan, CLS).

To monitor RLS experimentally, it is necessary to have systems that can recognise the mother and daughter cell at each of the divisions. On the other hand, to monitor CLS, it is necessary to have systems that distinguish, over time, which cells can re‐enter the cell cycle from those that cannot. This usually involves plating and colony‐forming unit counting.

Yeast has traditionally represented an amenable model to characterise ageing‐associated factors despite the fact that chronological and replicative lifespan experiments used to be low‐throughput, laborious and resource‐ and time‐consuming. Indeed, extensive research prior to the development of the technologies discussed here has identified a large number of evolutionarily conserved ageing‐related genes in yeast, such as TORC1 and sirtuin pathways, which also modulate lifespan in animal models (Kaeberlein, McVey & Guarente, 1999; Kapahi et al., 2010). At present, lifespan studies have massively benefited from HT pipelines to perform genome‐scale experiments, dealing with large numbers in parallel to guarantee global coverage and allow for robust statistical analyses (Coody & Hughes, 2018).

Recent HT experimental pipelines and new devices, in combination with pooled barcoded‐mutant collections and NGS, are unveiling novel genes and pathways that influence lifespan. Romila et al. (2021) decoded 3,206 mutants by barcode sequence from the latest version of the S. pombe gene‐deletion library to screen for CLS, identifying new long‐ and short‐lived mutants.

Along with known control genes that validate the system, they found up to 46 conserved but uncharacterised age‐related genes (orthogonally confirmed) including pathways such as iron–sulphur cluster formation, oestrogen action, and collagen biosynthesis (Romila et al., 2021). Another work from the same laboratory, by massive sequencing of quiescent cells, proposed a novel R‐loop‐mediated mechanism for the accumulation of chromosomal rearrangements that occur in non‐dividing aged yeast cells resembling those found in old brain cells of humans. They also identified genome hotspots associated with 3′ UTRs of highly transcribed genes with special enrichment in those targets of Scw1 (strong cell wall) RNA‐binding protein (Ellis et al., 2021).

Another recent study interrogating the barcoded S. cerevisiae deletion collection for extended RLS shows that class‐II histone deacetylases (HDAs) activity on histone 3 controls lifespan by epigenetic regulation of trehalose gene promoters. Defective HDA activity leads to increased trehalose levels that activate stress resistance responses and extend lifespan. The authors also showed that this mechanism is conserved, at least in Caenorhabditis elegans and Drosophila melanogaster (Yu et al., 2021).

Other methodological approaches that involve cross‐breeding of whole mutant collections and the HT replica‐pinning strategy allow us to enquire into more specific aspects of the cellular ageing process. Novarina et al. (2020) combined the yeast knockout (YKO) collection with an elegant genetic system previously developed in budding yeast, known as the Mother Enrichment Program (MEP).

MEP is an inducible oestradiol‐controlled system that specifically blocks the growth of the daughter cell of each division, but does not affect the proliferation of the mother cell, resulting in linear colony growth. Due to spontaneous mutations, MEP can be inactivated and the blocking daughter cells become insensitive to oestradiol, reverting to exponential colony growth. Therefore, colony size can be used as an indicator of mutagenic potential (Lindstrom & Gottschling, 2009).

Thus, using MEP ‘escaper’ formation as a readout for spontaneous mutagenesis events, Novarina et al. (2020) performed a genome‐wide screen aimed at identifying important genes to suppress the accumulation of spontaneous mutations in replicative ageing. Interestingly, they showed that deletion of PEX19 and PEX3 (both involved in peroxisome biogenesis and protein sorting) has no effect on mutagenesis in young cells, but causes an elevated accumulation of mutations specifically in aged cells (Novarina et al., 2020).

Microfluidic devices have also recently emerged as a relatively HT method for studying RLS. These consist of excavated slides, coupled to automated time‐lapse microscopy systems, where the mother cell is trapped and the number of divisions that each mother can generate can be monitored by video‐microscopy (Jo et al., 2015). In addition to its application in classical models, this system has opened the door to HT studies of RLS in pathogenic yeasts such as Cryptococcus neoformans.

This also has implications in pharmacology as ageing dynamics in these yeasts is related to virulence of infections and resistance to antifungals (Orner et al., 2019).

BIOMEDICAL APPLICATIONS: HIGH‐THROUGHPUT MODELLING OF HUMAN GENETICS AND DISEASE IN YEAST

Despite the evolutionary divergence between yeast and humans, estimated at one billion years (Sipiczki, 2000), a large majority of basic cellular functions and the genetic pathways that regulate them are highly conserved. Given the obvious importance of understanding the genetic basis of different diseases and the difficulty of studying them directly in humans and other animal models, for ethical and methodological considerations, budding yeast was initially used as a superb genetic model of our own biology in health and disease. However, in recent years, other evolutionarily different fission yeasts, such as S. pombe and S. japonicus, have emerged as models in impressive large‐scale studies (reviewed in Acs‐Szabo, Papp & Miklos, 2024), the findings of which can be extrapolated to human biology.

These studies can be classified by the type of methodological approach (reviewed in Laurent et al., 2016) and are outlined below.

Basic modelling

These are traditional studies that look for conditions in yeast (such as mutations, drugs, environment, etc.) that phenocopy either lack/gain of function or chemical/genetic interactions of the corresponding orthologous gene in humans. Dominant optic atrophy (DOA), for example, is a disease that causes blindness in humans. One of the molecular causes of DOA lies in the lack of function of the OPA1 gene.

This gene is essential for the mitochondrial dynamics of fusion, fission and mitophagy and the proper structural organisation of the cristae. To date there are more than 600 variants associated with human diseases (di Nottia et al., 2024). Aleo et al.

(2021) provide a good recent example of HTS approach for the modelling of the DOA. In this case, the I382M mutation in the human OPA1 gene is equivalent to the temperature‐sensitive I322M mutation in the yeast MGM1 orthologue gene. At restrictive temperature, cells carrying this allele show lethality or very limited growth on oxidisable carbon sources such as ethanol or glycerol.

In a search of the more than 1,600 compounds previously approved by the food and drug administration (FDA), they found six candidates able to rescue the growth defect of the mgm1‐I322M mutant. They then moved from yeast to human cell testing of the potential therapeutic drugs they identified. The authors show how some of these drugs also rescue relevant phenotypes in fibroblasts from DOA patients (Aleo et al., 2021; Gilea et al., 2021).

Additionally, yeast can be used not only as a model for genetic diseases, but also for studying fundamental cellular processes altered in the host cell after viral or prion infection (Sahaya Glingston et al., 2021; Acs‐Szabo et al., 2024).

Heterologous expression

Human genes without orthologues may also be expressed in yeast. If pathological alleles display a readable phenotype, they constitute good models in order to search for molecules that ameliorate or abolish such phenotype by tackling its molecular cause. An excellent early body of work exemplifies this approach applied to complex neurodegenerative disorders such as Parkinson's (PD) and Alzheimer's (AD) diseases.

The researchers first stablished budding yeast models expressing either human wild‐type and pathogenic alpha‐synuclein alleles or fragments of huntingtin exon 1 (HTTEX1) followed by poly(Q) tracts that generate aggregation and toxicity in yeast, as in human cells. They then screened thousands of cyclic peptides or hundreds of thousands of chemical compounds that restore yeast growth. Remarkably, the identified molecules were able to reverse pathological phenotypes in primary neuronal cultures, patient induced pluripotent stem cells (iPS)‐derived neurons and a PD nematode model.

Finally, by going back to lethal concentrations of these compounds in wild‐type yeast and searching for genetic conditions (mutations and over‐expression) that revert growth, they uncovered the molecular and cellular mode of action of these molecules (Kritzer et al., 2009; Khurana & Lindquist, 2010; Su et al., 2010; Tardiff et al., 2013, 2014; Chung et al., 2013).

Heterologous expression may also be tested for positive and negative genetic interactions within endogenous pathways. A good example of this strategy is the expression of deoxycytidine kinase (dCK) in S. cerevisiae to elucidate the differential interaction between anticancer drugs and gene networks. Budding yeast lacks the dCK orthologue and is therefore naturally resistant to the nucleotide analogues gemcitabine and cytarabine.

Santos et al. (2019) expressed this enzyme in the deletion collection in the presence of different concentrations of these drugs. They finely characterised the proliferation capacity by quantitative HT cell array phenotyping (Q‐HTCP).

Interestingly, by comparing the distinct patterns of gene–chemistry interactions for both drugs, they found clearly different genetic pathways despite their similar mode of action. Furthermore, by cross‐referencing the data with cancer pharmacogenomics databases, they found a correlation between drug‐sensitive deletions in yeast and the human homologues that are under‐expressed in gemcitabine/cytarabine‐sensitive cells. Conversely, yeast gene‐deletion suppressors identify human homologues that are overexpressed in drug‐sensitive cells.

This resource may then predict cancer cells' gene expression profiles that are causal in tumour‐specific response to cytotoxic agents (Santos et al., 2019).

Another variant of heterologous expression in yeast is deep mutational scanning. This involves analysing quantitatively and qualitatively the function of hundreds or thousands of allelic variants of any gene whose activity is measurable in some way. Seuma et al.

(2021) recently generated an expression library in S. cerevisiae of all possible nucleotide variants (>14,000) in the gene coding for amyloid beta (Aß) peptide, coupled to an ingenious detection system based on the aggregation of a yeast prion (Sup35) (Chandramowlishwaran et al., 2018). Plaques generated from mutants in Aß peptide are a hallmark of AD. As a proof of concept, they verified that all known human mutations responsible for an early manifestation of AD significantly accelerated the aggregation of Aß protein in yeast.

But in‐depth analysis of the results also revealed other previously unknown variants that affect Aß aggregation dynamics, identifying negatively charged residues as gatekeepers since their mutation severely increases aggregation capacity. Consistently, the introduction of negatively charged residues at other positions strongly reduces the aggregation capacity. The authors also described a clear bias towards the N‐terminal end of the protein amongst mutations that increase aggregation (Seuma et al., 2021).

Mirroring disease‐associated protein motifs

This logic of this approach is based on proteins with orthologues in humans with highly conserved motifs, in which substitutions associated with diseases can be mimicked. Viéitez et al. (2022) generated a collection of 474 phosphosite mutants in S. cerevisiae that were challenged in 102 different conditions.

Regarding those conserved in human cells, the severity of fitness loss is indicative of functional relevance (Viéitez et al., 2022).

By contrast, Ollodart et al. (2021) established a method to study continuous cultures in parallel. An interesting aspect of this type of approach, where samples can be taken over time, is that it allows addressing not only absolute and static phenotypes, but also dynamic phenotypes, such as the mutation rate in a continuous culture in a chemostat.

MutS Homolog (MSH2) is an enzyme in the mismatch repair pathway. Mutations in this gene are responsible for an elevated risk of many cancers, especially hereditary non‐polyposis colorectal cancer (Ollodart et al., 2021). The authors evaluated the mutation rate over time resulting from 185 different missense variants of the human MSH2 enzyme that they found in the ClinVar repository (Landrum et al., 2014), which were recreated at the counterpart sites of the S. cerevisiae protein.

As a control, they were able to recapitulate 100% of the 28 known pathogenic variants, but they also identified at least 50 variants of unknown clinical significance that raise the mutation rate in comparison with the wild‐type allele.

Entire gene substitution

There are numerous examples where a yeast gene can be complemented by its human counterpart. These studies have important implications for evolutionary studies. In cases when their function can be quantified in some way (growth rate, metabolic activity, biosensors, cell phenotype, drug sensitivity, etc.), it is possible to probe the activity of alleles naturally found in the human population to correlate polymorphism with a disease's expressivity in humans (Hamza et al., 2015; Yang et al., 2017).

On the other hand, when the orthology ratio is greater than 1:1, there are interesting approaches in which a given yeast gene is systematically replaced by each of the representatives of its orthologous group in humans (in some cases this can comprise more than 20 genes). Garge et al. (2020) systematically humanised major elements of the yeast cytoskeleton by replacing corresponding yeast genes with their orthologous human counterpart families.

They found that 13 out of 50 genes tested can perform essential functions in yeast. Within each family, this work shows that more than one member can complement the essential function for growth of each yeast gene. But they also observed complementation, to varying degrees, of non‐essential functions of processes involving the cytoskeleton such as cell morphology, sporulation, mating, meiosis, and cytokinesis (Garge et al., 2020).

Whole‐pathway implementation or replacement

Given the modularity of biological functions at the genetic level, it is in principle possible to extend the above logic to several elements or even to entire genetic pathways. Amorosi et al. (2021) measured, respectively, the activity and abundance of over 6000 missense variants of the cytochrome P450 family 2 subfamily C member 9 (CYP2C9).

The activity of this enzyme depends on accessory proteins and is responsible for the metabolism of a large number of drugs, including warfarin, so its polymorphism is of immense importance in the field of pharmacogenomics. The authors generated activity and abundance scores for a barcoded library of 6142 missense variants in a humanised yeast strain also expressing human CYP accessory proteins [cytochrome P450 reductase (CPR) and b5] to improve sensitivity. The data show that two thirds of the variants exhibit a significant reduction in activity and half of them were correlated with a decrease in abundance.

On the other hand, residues involved in heme coordination and binding are critical for activity but not for abundance (Amorosi et al., 2021). These approaches provide structural insights and clinical impacts for human protein polymorphisms and can be used as platforms for therapeutic testing.

All these, and numerous other examples where yeast is humanised following any of the above approaches (see Laurent et al., 2016; Cervelli & Galli, 2021; Moesslacher, Kohlmayr & Stelzl, 2021), show the power of yeast for understanding and fighting human disease.

BIOTECHNOLOGICAL APPLICATIONS: DRUG DISCOVERY AND PRODUCTION OF COMPOUNDS

The systematic testing of different molecules (or combinations of them) in different genetic backgrounds with variations across the genome is known as chemical genomics. These studies mainly reveal the molecular mode of action of different compounds and their impact upon genetic pathways. This is undoubtedly one of the fields that has benefited most from mass screening in the discovery of compounds of interest.

Indeed, this was one of the driving forces in the conception and development of HTS and ultra‐HTS experimentation more than two decades ago (Hoon et al., 2008; Ho et al., 2009; Ohnuki et al., 2010). The enormous combinatorics offered by the integration of robotics, informatics, molecular biology, genetics, chemistry and biochemistry allows the almost unlimited design of screens for thousands of synthetic and/or natural molecules in thousands of combinations, concentrations and genetic backgrounds, all for the identification of biochemical activities of biomedical or biotechnological interest. As mentioned above, the evolutionary conservation of multiple genetic pathways and the versatility in the ‘humanisation’ of yeast at various levels, coupled with the number of phenotypes that can be used as biosensors, facilitate rational designs based on genetic modifications in yeast that simulate molecular malfunctions underlying human genetic diseases (reviewed in Denny & Steel, 2015; Denny, 2018).

But they can also be applied to toxicity studies with environmental and public health implications (De La Rosa et al., 2017; Kumar et al., 2024).

Since large collections of yeast mutants have become available, there have been numerous studies in which yeasts have been exposed to different drugs. Countless examples include conazole (Guan et al., 2020), genotoxic agents (Guan et al., 2022), the environmental contaminants trichloroethylene (De La Rosa et al., 2017) and formamide (Hoyos‐Manchado et al., 2017) and pharmaceutical contaminants (Ogbede, Giaever & Nislow, 2021). Moreover, yeast offers a model to study processes related to the pharmacodynamics of new drugs.

Dias et al. (2021) recently identified the import and export pathways of different compounds using a collection of more than 14,000 combinations of double mutants of non‐essential membrane transporters. Such studies may be of great interest in the design of new molecules that achieve optimal uptake and/or retention in the tissues of interest (Dias et al., 2021).

In another multidimensional assay, Lee et al. (2014) examined the effect on fitness of 3250 compounds in 5900 different diploid strains, classified into two groups: the sensitivity of missing genes considered essential was assessed in 1100 deletions in heterozygosis by haploinsufficiency profiling, while non‐essential genes were analysed in about 4800 homozygous deletions (homozygous profiling). More than 300 of these compounds showed an effect on 121 genes classified into 45 cellular response signature networks (Lee et al., 2014).

However, perhaps the most direct and intuitive application of HTS designs in yeast compound screening is the search for antifungals. Among the phylum Ascomycota, there are many yeasts and fungi that are pathogenic in humans, as well as in animals and plants of economic interest. Candida and Aspergillus species, for example, are major causes of opportunistic fungal infections in humans.

These infections often share the characteristic of associating their infectivity, and resistance to treatment, with filamentation and biofilm formation. It is therefore necessary to have effective treatments against these infections. This approach is as simple as looking for molecules, compounds or combinations of them that cause death in the species of interest but are harmless to the host.

Therefore, the design of chemical screens that inhibit these processes offer promising antifungal candidates (Pierce, Saville & Lopez‐Ribot, 2014; Hoffman, 2022; Kelty, Miron‐Ocampo & Beattie, 2024; Gutierrez‐Perez et al., 2024).

On the other hand, the combination of biosensors, gene collections and suitable detection systems also allows the search for specific inhibitors or activators. A recently developed strategy, generically known as yeast surface display, uses genetically modified yeast cell wall proteins as anchors for enzymes of interest. This converts the cell wall into a living reactor for HT searches (reviewed in Lopez‐Morales et al., 2023) in order to optimise industrial processes for the biosynthesis of compounds of biotechnological interest.

Baumann et al. (2021) combined a fluorescent sensor with an overexpression gene library and FACS to find that the combined overexpression of the KCS1 and FSH2 genes increases the production of octanoic acid by up to 55%. This compound has antimicrobial capabilities and it is a precursor for the production of biofuels (Baumann et al., 2021).

In a similar scenario, deletion of PAD1 and FDC1 confers resistance in S. cerevisiae to gamma valerolactone (GVL), an effective compound in the degradation of lignocellulosic biomass for biofuel production, which is normally toxic to fermentative microbes. This opens avenues for xylose fermenting yeast specifically tailored to GVL‐produced hydrolysates (Bottoms et al., 2018). Using a different approach, Cribari et al.

(2023) designed an ultra‐HT FACS‐sorted biosensor to identify activity‐boosting mutations, among a library of more than ten million enzyme variants, that substantially increased the kinetics of degradation of solid polyethylene terephthalate (PET) films by the leaf and branch compost cutinase.

Another strategy in the field of optimising yeasts as bioreactors is directed evolution (Section IV.2). This rationale is not based on gene inactivation or overexpression, but on the search for protein evolution in continuous cultures to produce higher yields and/or products of interest. Wellner et al.

(2021) have recently paired OrthoRep to yeast surface display to achieve rapid generation of antibodies. Jensen et al. (2021) show another direct application of OrthoRep by iteratively mutating two subunits of protocatechuate decarboxylase AroY from Klebsiella pneumoniae in S. cerevisiae to achieve 13‐fold higher titres of cis‐cis‐muconic acid (CCM) compared to controls.

This compound is used in the production of bioplastics, coatings and nylon (Jensen et al., 2021). A good example of different but convergent HTS strategies in this field is a recent study where Pyne et al. (2023) also pursue the improvement of CCM production by the screening and genetic engineering of non‐conventional yeasts.

These yeasts are likely to become increasingly prominent because they contribute a great diversity of metabolic pathways. In some instances, they are naturally superior to S. cerevisiae in their resistance to toxic precursors or in the production yield itself. Pyne et al.

(2023) isolated and engineered a Pichia occidentalis strain that surpasses all metrics achieved using S. cerevisiae in the production of CCM. Another example of the potential of diverse yeasts for valuable compound production is the screen of 1,434 non‐conventional yeasts for microbial oil production by utilising short‐chain fatty acids found in organic waste. Two strains from Yarrowia lipolytica and Pichia manshurica were found to display significant ability to produce lipids compared to reference strains (Žganjar et al., 2024).

CONCLUSIONS

(1)The crossover between mechanical engineering, chemistry, genetics and computer science now offers an unprecedented ability to address fundamental and large‐scale biological questions that were unthinkable some years ago.(2)Using yeast as a biological model is having an immediate impact, both in fundamental science and in applications to diverse fields. It has been used in various ways, from understanding and treating diseases of high social impact, to new methods of production and degradation of compounds of interest capable of generating added value in pharmaceuticals, bioindustry and environmental protection.(3)New technical developments in the near term in the disciplines involved will move the experimental capacity from high to ultrahigh at all levels.(4)It is foreseeable that the explosion of artificial intelligence will contribute to generating powerful interactive databases, resources and tools where HTS results from different experiments in time, space and species may be integrated. This could allow for deeper – or different – analyses that make the most of interpretation to reveal unexpected connections and applications.(5)Machine learning will likely help to rationalise experimental design based on previous modelling and thus improve the costs, efficiency and significance of this experimentation.

The only bottleneck in this field will be the art of experimental design.

Related articles

PLOS

Estimating rates of SARS-CoV-2 lineage spread using graph spectra from genetic sequence data

The modern enormous scale of viral genomic data, as exemplified by SARS-CoV-2, presents unique opportunities for epidemic inference and real-time monitoring. A key challenge is to wade through the extremely large stream of sequence data in order to identify small subsets of sequences that may represent rapid population growth of newly emerged lineages. To address this challenge, we developed a new approach that combines techniques in graph theory and machine learning. We represent distances betw...

PubMed

Leveraging artificial intelligence for livestock genetics and breeding.

The rapid development of artificial intelligence (AI) is transforming modern society including animal genetics and breeding which is essential for meeting global nutritional needs in supplying milk, meat, and eggs among the most complete sources of protein and micronutrients. Although animal breeding has a long history of using data-driven prediction methods, the application of modern AI approaches remains at a relatively early stage and shows substantial potential for wide-ranging applications...